Download CPSC 444 Artificial Intelligence What Is AI?

Survey
yes no Was this document useful for you?
   Thank you for your participation!

* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project

Document related concepts

Embodied cognitive science wikipedia , lookup

Artificial intelligence in video games wikipedia , lookup

Knowledge representation and reasoning wikipedia , lookup

Technological singularity wikipedia , lookup

AI winter wikipedia , lookup

Philosophy of artificial intelligence wikipedia , lookup

Ethics of artificial intelligence wikipedia , lookup

Intelligence explosion wikipedia , lookup

Existential risk from artificial general intelligence wikipedia , lookup

History of artificial intelligence wikipedia , lookup

Transcript
What Is AI?
CPSC 444
Artificial Intelligence
• “The main unifying theme is the idea of an
intelligent agent. We define AI as the study of
agents that receive percepts from the
environment and perform actions.”
– Stuart Russell & Peter Norvig
[authors of a well-known AI textbook]
CPSC 444: Artificial Intelligence • Spring 2017
What Is AI?
What Is AI?
• “The science of making machines do things
that would require intelligence if done by
humans.”
– Marvin Minsky
• “Artificial Intelligence (AI) is the study of solutions for
problems that are difficult or impractical to solve with
traditional methods.”
– ACM/IEEE CS2013 Curriculum Guidelines
3
[pioneer/founder in AI and Turing Award winner]
• “The use of computer programs and
programming techniques to cast light on the
principles of intelligence in general and
human thought in particular.”
– Margaret Boden
• “Artificial Intelligence is whatever hasn't been done yet.”
– common misquoting of Tesler's Theorem
[cognitive scientist]
CPSC 444: Artificial Intelligence • Spring 2017
2
CPSC 444: Artificial Intelligence • Spring 2017
4
Areas
Areas
• communicating, perceiving, acting
• problem solving
– natural language processing
– perception / vision
– search
– constraint satisfaction
– deduction and reasoning
• use input from sensors e.g. cameras, microphones, sonar, …
• applications include speech recognition, facial recognition, object
recognition
– robotics
• planning
• object manipulation, navigation, mapping, motion/path planning, …
– find a sequence of actions that lead to a particular goal
– classical planning assumptions
•
•
•
•
•
•
• knowledge representation
unique known initial state
durationless actions
deterministic actions
can only take one action at a time
can accurately predict the state of the world after an action is taken
single agent
– many problems require knowing about the world
– storing information about
•
•
•
•
•
– beyond classical planning
CPSC 444: Artificial Intelligence • Spring 2017
5
objects, properties, categories of objects, relationships between objects
situations, events, states, and time
causes and effects
what we know about what other people know
…
CPSC 444: Artificial Intelligence • Spring 2017
Areas
Areas
• learning
• other types of intelligence
– program learns from experience if it does better on a task for
having had the experience
7
– social intelligence – recognize, interpret, process, and simulate
feelings and emotions
– creativity
– general intelligence – can do any intellectual task a human can
(not limited to a specific domain)
• artificial life
– studies processes of life and systems related to life through
simulations and models
– using computers to tell us something about non-humans and
humans
CPSC 444: Artificial Intelligence • Spring 2017
6
CPSC 444: Artificial Intelligence • Spring 2017
8
Course Topics
Course Materials
• reactive agents and simple decision making
http://math.hws.edu/bridgeman/courses/444/s17/
• smarter agents I
– planning
– [reasoning]
– problem solving (via search)
• smarter agents II
– evolution
– machine learning
• philosophical and ethical issues, ramifications
– what is intelligence and how can we identify it?
– will AI destroy the world?
CPSC 444: Artificial Intelligence • Spring 2017
9
CPSC 444: Artificial Intelligence • Spring 2017
Prerequisites
Course Materials
• C- or better in CPSC 327 or CPSC 329
• no textbook to purchase – readings will be posted
Expectations and assumptions –
• primary programming languages will be Java and
Processing
• comfortable with using programming as a tool
– fluent in Java syntax and semantics
– comfortable writing classes and working with objects
– familiar with Java Collections classes and when to use Array,
Stack, Queue, PriorityQueue, Map
– can translate ideas into code
– can decide on program organization (classes and methods)
12
• all of the necessary software is available on the lab
machines in Rosenberg 009 and Lansing 310
– course information page has information on acquiring Java,
Eclipse, Fugu/WinSCP, and Processing if you want to set up
your own computer (optional)
• you will ask questions or figure things out yourself when
you don't know something
CPSC 444: Artificial Intelligence • Spring 2017
11
CPSC 444: Artificial Intelligence • Spring 2017
13
Course Schedule
check here for readings, assignments,
handouts, examples from class, etc
Collaboration Policy
Discussing ideas with and getting debugging help
from other students is OK.
Working with others to produce a solution that
everyone hands in is not OK (except for group
assignments)
– even if you contribute to the creation of that solution
– even if you write/type up the solution yourself
– even if you make some modifications later
– this includes working side by side, frequently consulting
as you each write down things on your own computer
CPSC 444: Artificial Intelligence • Spring 2017
14
Important Schedule Notes
CPSC 444: Artificial Intelligence • Spring 2017
16
Being Successful
Stay caught up – review material promptly after
class and ask questions when you have them.
Plan sufficient time for assignments – start early.
– projects (but not homeworks) will be accepted late, but it
is easy to fall into a habit of being late and that will
impact your grade
Utilize office hours.
– backgrounds vary, so the expectation is that you will ask
questions (or figure things out yourself) rather than
having extensive detail in assignment handouts
Attend class.
CPSC 444: Artificial Intelligence • Spring 2017
15
CPSC 444: Artificial Intelligence • Spring 2017
17
“Gestation”
1943-1955
• Warren McCulloch and Walter Pitts
– proposed a network of artificial neurons, and
its capability of computing any computable
function
• Marvin Minsky and Dean Edmonds
– built a neural network computer (SNARC)
simulating a rat learning to escape a maze
• Alan Turing
– “Computing Machinery and Intelligence”
introduces the Turing Test, machine learning,
genetic algorithms, and reinforcement
learning
CPSC 444: Artificial Intelligence • Spring 2017
18
20
CPSC 444: Artificial Intelligence • Spring 2017
Foundations of AI
“Birth”
• the idea that thinking is symbolic reasoning
• a 2-month workshop at Dartmouth brought together 10
researchers in automata theory, neural networks, and the
study of intelligence
– philosophers: Hobbes (1588-1679), Descartes (1596-1650),
Pascal (1623-1662), Spinoza (1632-1677), Leibniz (1646-1716)
– first use of “artificial intelligence”
• advances in mathematical logic
– Turing (1912-1954) – Turing machine
– Church (1903-1995) – λ-calculus
– Church-Turing thesis: Any function which has an algorithm can
be computed using a Turing machine. (paraphrased)
• the symbolic era (1950s – early 1980s)
– a set of knowledge
– a reasoning algorithm to manipulate those symbols to represent
problem solutions or new knowledge
• provided a link between the process of mathematical deduction and what
a mechanical symbol-manipulation device could do
– work focused on
•
•
•
•
• the invention of computers
– Babbage (1792-1871) – Analytical Engine
– 1940s: Z3 (Germany), ENIAC (US), Colossus (Britain)
CPSC 444: Artificial Intelligence • Spring 2017
1956
19
reasoning (based on search)
knowledge representation
expert systems
tradeoff of knowledge vs search
CPSC 444: Artificial Intelligence • Spring 2017
21
“The Golden Years”
1950s-1974
• lots of success in limited domains, breaking the
establishment view that “a machine can never do X”
“The Golden Years”
1950s-1974
Lots of optimism –
– Allen Newell and Herbert Simon
• “within ten years a digital computer will be the world's
chess champion” [Simon & Newell 1958]
• “within ten years a digital computer will discover and
prove an important new mathematical theorem” [Simon &
• Logic Theorist (1956)
• General Problem Solver (1957)
Newell 1958]
– Herbert Gelernter: Geometry Theorem Prover (1958)
• “machines will be capable, within twenty years, of doing
any work a man can do” [Simon 1965]
• “Within a generation … the problem of creating 'artificial
intelligence' will substantially be solved.” [Minsky 1967]
• “In from three to eight years we will have a machine with
the general intelligence of an average human being.”
– John McCarthy: proposal for Advice Taker (1959)
• intended to demonstrate common sense - “a program has
common sense if it automatically deduces for itself a sufficiently
wide class of immediate consequences of anything it is told and
what it already knows” [McCarthy 1959]
– Arthur Samuel: checkers program which learned to play
at a strong amateur level (1952-1962)
22
CPSC 444: Artificial Intelligence • Spring 2017
“The Golden Years”
[Minsky 1970]
1950s-1974
CPSC 444: Artificial Intelligence • Spring 2017
24
Reality Intrudes
• lots of grand promises made, but real progress was much
slower
– other progress
• in microworlds problems: e.g. calculus I integration problems,
geometric analogy problems found in IQ tests, algebra story
problems, blocks world
• in neural networks
– techniques that worked in limited domains fall apart on more
general problems
• difficulties
– the solution of many problems requires more than just syntactic
manipulations – background knowledge is important
• John McCarthy created Lisp (1958)
• but common sense knowledge is too vast
– intractable problems and combinatorial explosion
• 1971: Cook and Levin prove boolean satisfiability is NP-complete
• 1972: Karp proves 21 problems to be NP-complete
– insufficient computing power
– some basic structures were too simple to generate intelligent
behavior
• e.g. very simple neural networks can't be trained for more complex tasks
CPSC 444: Artificial Intelligence • Spring 2017
23
CPSC 444: Artificial Intelligence • Spring 2017
25
AI Winter
1974 – 1980
• a decline in many areas of AI because of failure to deliver
on grand promises
AI Spring
• AI becomes an industry
1980 – 1987
1980-
– first successful commercial expert system R1 helped configure
orders for new computer systems at DEC, eventually saving the
company $40 million per year
– companies building expert systems, vision systems, robots, and
related hardware/software
– funding dried up
• influential book Perceptrons [Minsky and Papert 1969] halts
most work on neural networks for 10 years
• not a complete drought
– expert systems
• e.g. MYCIN for diagnosing blood infections
– general themes
• knowledge-based systems
– limited domain of an expert system keeps the knowledge requirements
tractable
• importance of domain knowledge in understanding natural language
• issues of knowledge representation
26
CPSC 444: Artificial Intelligence • Spring 2017
28
CPSC 444: Artificial Intelligence • Spring 2017
Expert Systems
AI Spring
Expert systems diagnose and provide expert advice.
Applications include recommendations for insurance policies and
mortgages, optimizing patient care, disaster response procedures.
• resurgence and expansion of many ideas that fell out of
favor during the AI winter
– return of neural networks
1980 – 1987
1986-
• new algorithms and models
– larger databases possible
• Cyc (begun 1984) – an attempt to encode an average person's
commonsense knowledge
• adoption of the scientific method
1987-
– more common to build on existing theorems than create new
ones
– base claims on rigorous theorems or experimental evidence
rather than intuition
– demonstrate relevance on real-world applications rather than toy
examples
CPSC 444: Artificial Intelligence • Spring 2017
http://www.lpa.co.uk/wfs_dem.htm
27
CPSC 444: Artificial Intelligence • Spring 2017
29
Second AI Winter
1987 – 1993
• collapse of the market for specialized AI hardware
• oldest expert systems started to become too expensive to
maintain
Modern AI
1993 – present
• AI techniques incorporated into larger systems
– include aspects of intelligent behavior to achieve tasks
– e.g. scheduling application or automatic bidding system using
agents representing users to negotiate according to individual
constraints
– e.g. autonomous agents to control cars, spacecraft
• results once again fell short of high expectations
– more funding cuts
• AI techniques have lots of applications
– e.g. fuzzy logic and fuzzy control systems → camera auto-focus,
antilock brake systems
– e.g. collaborative filtering → product recommendations
– e.g. clustering techniques → organizing search results
“…once something becomes useful enough and common enough it's
not labeled AI anymore."
– Nick Bostrom [philosopher]
30
CPSC 444: Artificial Intelligence • Spring 2017
Modern AI
• emergence of intelligent agents
1993 – present
Modern AI
1995-
32
CPSC 444: Artificial Intelligence • Spring 2017
1993 – present
• availability of very large data sets
– legitimizes focusing on isolated problems
– AI algorithms making it out into other applications
2001-
– data can supplant algorithm in some ways
– e.g. “plant” - flora or factory?
• can learn to high accuracy from dictionary definitions of the two senses
and a very large corpus of unannotated text
– e.g. filling in gaps in a photo
• poor performance with 10,000 photos but excellent performance with
2,000,000
CPSC 444: Artificial Intelligence • Spring 2017
31
CPSC 444: Artificial Intelligence • Spring 2017
33
Modern AI
1993 – present
• computers continue to get faster and
more powerful
– 1997: IBM's Deep Blue beats reigning
world chess champion
– 2005: Stanford's Stanley drives 131 miles
across the desert to win the DARPA Grand
Challenge
– 2007: CMU's Boss drives 55 miles in an
urban environment
– 2011: IBM's Watson defeats Ken Jennings
and Brad Rutter in Jeopardy!
CPSC 444: Artificial Intelligence • Spring 2017
34